An Efficient Renewable Power Prediction Model Using IGOA-based Enhanced BiLSTM Deep Neural Network for Microgrid Smart Energy Management

Abstract

Accurate forecasting of renewable energy generation and operational variables is a critical requirement for effective smart energy management in microgrid systems. This paper presents an integrated forecasting framework based on a Bidirectional Long Short-Term Memory network optimized using an Improved Grasshopper Optimization Algorithm (IGOA-BiLSTM). The proposed method exploits the bidirectional learning capability of BiLSTM to model complex nonlinear temporal dependencies, while an enhanced Grasshopper Optimization Algorithm is introduced to optimally tune the network’s hidden layer size. The IGOA incorporates a dynamic inertia weight updating mechanism and an adaptive triangular mutation strategy to improve exploration–exploitation balance, enhance convergence stability, and avoid premature convergence. The model is evaluated using real-world hourly data from the PJM West Zone, including photovoltaic power, wind turbine power, load demand, and day-ahead electricity prices. Experimental results demonstrate that the proposed IGOA-BiLSTM achieves superior forecasting accuracy compared with conventional LSTM and state-of-the-art hybrid models. Specifically, the proposed method attains coefficients of determination (R²) of 0.99 for photovoltaic power, wind power, load demand, and day-ahead price forecasting. For photovoltaic and wind power prediction, RMSE values of 0.019 and 0.015 are achieved, respectively, with corresponding MSE values as low as 0.0003 and 0.0002. In load and day-ahead price forecasting, the model achieves RMSE values of 0.019 and 0.009, respectively. These results confirm the robustness, high accuracy, and computational efficiency of the proposed framework, making it well suited for practical microgrid energy management, renewable energy planning, and electricity market decision-support applications.